Claims in connection with the kind of knowledge to be captured

1. With sufficiently broad coverage of linguistic phenomena, we can use natural language interaction as a means of accessing the content of a high performance knowledge base. This is has been demonstrated to be practical today using the START system.

2. There is significant overlap between the core ontologies needed for HPKB and the information needed for the lexicon of our natural language system. Much of this information has already been studied and accumulated by computational linguists.

3. The Web is already a knowledge base, it's simply one without indexing structure (particularly so for the computational capabilities reachable through the Web). The rapid growth of the Web is attributable to its decentralized, anarchic character but so is its lack of structure.

   We will make the Web behave like a real knowledge base without losing the distributed, anarchic character that has been its strength. The core of this claim is that we can provide effective indexing for large volumes of assets available through the Web by using short textual annotations which summarize the content or capability available at each URL of interest. Furthermore, because the medium of description is natural language, any provider of capability or content can not only develop the content but can easily contribute to its indexing. This is equally true for textual pages, graphical elements, video segments or computational capabilities; natural language is the lingua franca.

4. The Web has provided a new context in which all online information is potentially part of one's individual library. This is normally thought of in terms of ``passive information,'' but importantly, active information can be part of that global library as well: a URL can as easily point to a source of expertise as to a document.

5. Capturing active ``know-how'' information is as important to high performance knowledge bases as is capturing static information. Often know-how is best expressed as high level procedures in which the implementation of the major steps invokes problem solving behavior available in the underlying knowledge representation system and its associated problem solvers. Our approach provides a natural means for capturing such know-how because know-how is easily described in natural language and our system will coerce English utterances into executable code for a variety of problem solving interpreters.

6. There is a representation level used by natural language systems in which the utterance has been syntactically reduced to its most primitive structural components but in which no semantic reduction has been performed. This level can serve as a unifying interlingua between representations specialized to the semantics of particular domains. Our natural language system will then serve as a universal query system which can answer many questions on its own but which can also encompass the reasoning power of domain specific problem solvers. Similarly, our natural language system will serve as a universal front end for acquiring both general purpose knowledge which it manages itself and for information intended to enrich its associated domain specific representation and reasoning systems. Thus, our natural language system can serve as a unifying agent leading to a confederation of knowledge-rich systems.